This DevTalks webinar explores how SambaNova works with CrewAI. The demo showcases the power of SambaNova in CrewAI’s agentic framework and enables viewers to immediately build their own AI agents crew that interacts together to solve real-world problems.
This DevTalks webinar introduces developers to designing effective AI agents using CrewAI, with a focus on how high-performance inference infrastructure enables more powerful agentic systems. Justin Woo from SambaNova provides a brief introduction to the SambaNova SN40L RDU chip, and why inference speed and efficiency matter as AI applications evolve from simple chatbots to complex, multi-agent workflows that generate far more tokens and coordinate across multiple models. He demonstrates SambaNova’s inference performance with a live side-by-side comparison against ChatGPT, showing dramatically faster time-to-first-token and overall generation speed while maintaining accuracy.
Shane from CrewAI then framed the evolution of AI applications in three phases: single chatbots, domain-specific assistants and copilots (often powered by RAG), and today’s agentic workflows. In this latest phase, multiple specialized agents collaborate to handle complex, repeatable business processes with a higher degree of autonomy. He emphasized that real value comes not just from orchestration, but from being able to build, observe, optimize, and scale agents across an organization — supported by features like visual editors, tracing, guardrails, human-in-the-loop feedback, and centralized agent and tool repositories. A case study from General Assembly illustrated how mapping agents to real human roles reduced content development time by 90% and helped reduce team burnout.
The live demo showed how CrewAI’s visual studio can generate a full multi-agent workflow from a detailed natural-language prompt, automatically defining agent roles, tasks, and backstories. Justin and Shane highlighted how developers can iteratively refine these agents, swap models per agent based on task needs, replay individual steps, and integrate or even auto-generate custom tools. Throughout, they stressed that agentic systems are inherently “chatty” and token-intensive, making fast, cost-efficient inference — like SambaNova’s — especially valuable. The webinar concluded with Q&A, practical use cases (from competitive research to scheduling and email automation), and encouragement for attendees to experiment hands-on, mix and match models, and start building agentic workflows today.